Main content area

Call it a “nightshade”—A hierarchical classification approach to identification of hallucinogenic Solanaceae spp. using DART-HRMS-derived chemical signatures

Beyramysoltan, Samira, Abdul-Rahman, Nana-Hawwa, Musah, Rabi A.
Talanta 2019 v.204 pp. 739-746
Atropa, Brugmansia, Datura, Mandragora, algorithms, atropine, biomarkers, forensic sciences, landscaping, mass spectrometry, models, ornamental plants, scopolamine, seeds, species identification
Plants that produce atropine and scopolamine fall under several genera within the nightshade family. Both atropine and scopolamine are used clinically, but they are also important in a forensics context because they are abused recreationally for their psychoactive properties. The accurate species attribution of these plants, which are related taxonomically, and which all contain the same characteristic biomarkers, is a challenging problem in both forensics and horticulture, as the plants are not only mind-altering, but are also important in landscaping as ornamentals. Ambient ionization mass spectrometry in combination with a hierarchical classification workflow is shown to enable species identification of these plants. The hierarchical classification simplifies the classification problem to primarily consider the subset of models that account for the hierarchy taxonomy, instead of having it be based on discrimination between species using a single flat classification model. Accordingly, the seeds of 24 nightshade plant species spanning 5 genera (i.e. Atropa, Brugmansia, Datura, Hyocyamus and Mandragora), were analyzed by direct analysis in real time-high resolution mass spectrometry (DART-HRMS) with minimal sample preparation required. During the training phase using a top-down hierarchical classification algorithm, the best set of discriminating features were selected and evaluated with a partial least square-discriminant analysis (PLS-DA) classifier to discriminate and visualize the data. The method yields species identity through a class hierarchy, and reveals the most significant markers for differentiation. The overall accuracy of the approach for species identification was 95% and 96% using 100X bootstrapping validation and test samples respectively. The method can be extended for the rapid identification of an infinite number of plant species.